Stable Target Field for Reduced Variance Score Estimation in Diffusion Models

TL;DR

This paper introduces a stable target field method for diffusion models that reduces training target variance by using a reference batch, leading to improved image quality, stability, and faster training, achieving state-of-the-art results on CIFAR-10.

Contribution

The paper proposes a novel stable target calculation using a reference batch to reduce variance in diffusion model training, enhancing performance and stability.

Findings

Reduced covariance of training targets improves model stability.

Enhanced image quality and training speed across datasets.

Achieved state-of-the-art FID score of 1.90 on CIFAR-10.

Abstract

Diffusion models generate samples by reversing a fixed forward diffusion process. Despite already providing impressive empirical results, these diffusion models algorithms can be further improved by reducing the variance of the training targets in their denoising score-matching objective. We argue that the source of such variance lies in the handling of intermediate noise-variance scales, where multiple modes in the data affect the direction of reverse paths. We propose to remedy the problem by incorporating a reference batch which we use to calculate weighted conditional scores as more stable training targets. We show that the procedure indeed helps in the challenging intermediate regime by reducing (the trace of) the covariance of training targets. The new stable targets can be seen as trading bias for reduced variance, where the bias vanishes with increasing reference batch size. Empirically, we show that the new objective improves the image quality, stability, and training speed of various popular diffusion models across datasets with both general ODE and SDE solvers. When used in combination with EDM, our method yields a current SOTA FID of 1.90 with 35 network evaluations on the unconditional CIFAR-10 generation task. The code is available at https://github.com/Newbeeer/stf